Potential Vs. Realized Savings Under Automatic Enrollment

Trends and Issues July 2018

Potential vs. realized savings under automatic enrollment

I. Overview John Beshears, Previous research has documented the powerful impact that automatic enrollment Harvard University has on retirement savings outcomes. When a savings plan’s default—the option that and NBER is implemented on behalf of any employees that do not actively elect an alternative James J. Choi, option—is changed from not participating in the plan to contributing a positive fraction Yale University of pay to the plan, the proportion of employees contributing to the plan increases and NBER dramatically, and many employees who would otherwise have not participated begin David Laibson, to accumulate plan balances (Madrian and Shea, 2001; Choi, et al., 2002 and 2004; Harvard University Beshears, et al., 2008). Even though automatic enrollment increases plan contributions and NBER for many employees, the ultimate impact on the accumulation of plan balances is Brigitte C. Madrian, unclear. Argento, Bryant and Sabelhaus (2015) have documented that many households Harvard University make substantial withdrawals from their defined contribution accounts well before and NBER reaching retirement age, a phenomenon known as “leakage” (because balances are “leaking” out of accounts). They find that among households under the age of 55, each dollar contributed to a 401(k) plan or similar tax-advantaged retirement account is offset by approximately $0.40 in pre-retirement taxable withdrawals. This high rate of leakage raises the possibility that the positive effect of automatic enrollment on savings plan contributions may be offset in whole or in part by subsequent pre- retirement withdrawals, leaving a reduced long-term net impact (or no net impact) of automatic enrollment on retirement assets.

Any opinions expressed herein are those of the authors, and do not necessarily represent the views of TIAA, the TIAA Institute or any other organization with which the authors are affiliated. We studied the effect of automatic enrollment on savings potential increase in savings from automatic enrollment. plan loans and withdrawals and their implications for The net effect is that automatic enrollment increases the evolution of retirement plan balances over time retirement system balances by 4-5% of first year pay by examining the experience of a large Fortune 500 eight years after hire. These results mask substantial company in the financial services sector that introduced differences across those who remain employed at the automatic enrollment at a 2% default contribution rate firm versus those who separate. Among those who for all employees hired on or after July 1, 2005. Our remain employed, leakage offsets relatively little of the empirical strategy compares savings plan outcomes for incremental savings generated by automatic enrollment employees hired in the 12 months after the introduction at low levels of tenure. As tenure increases, so does the of automatic enrollment to those for employees hired extent to which leakage offsets the savings increases in the 12 months prior. We restrict our analysis to from automatic enrollment, and eight years after hire, those employees in both cohorts who remained at leakage, primarily in the form of plan loans, offsets the firm for at least one year and then follow these 9-27% of the potential increased savings. In contrast, for two cohorts for up to eight years after they joined the employees who separate, leakage, primarily in the form firm. We first examine outcomes directly observable non-rollover withdrawals, offsets more than half of the in administrative data: savings plan participation, potential incremental savings from automatic enrollment contributions, balances, outstanding loans, and whether at low levels of tenure. Although this rate of offset plan withdrawals are rolled over into another qualified declines with time since hire for separated employees, at savings plan or not. We then project the potential impact eight years it still exceeds 40%. Overall, while automatic that automatic enrollment could have on retirement enrollment results in a net increase in retirement system savings accumulations if there were no plan leakage and balances, preretirement leakage significantly limits its decompose that amount into several component parts— potential impact. retirement plan balances, outstanding loan balances, rollovers into other qualified plans, and non-rollover II. Data and methodology withdrawals—to quantify the extent to which leakage reduces retirement asset accumulation overall, and the We studied a large U.S. Fortune 500 company in the incremental asset accumulation induced by automatic financial services sector. Table 1 summarizes the enrollment in particular. relevant features of the retirement savings plan at this firm. To examine the extent to which retirement plan Consistent with previous research, we find that savings leakage offsets the increased savings that results plan participation at the firm we study is significantly from automatic enrollment, we analyzed employee- higher for the post-automatic enrollment cohort in level data on a single client firm of a large U.S. benefits the first few years after being hired, as is the average administrator. The data consist of a series of year-end fraction of pay contributed to the plan; conditional on cross sections containing demographic and employment- plan participation, however, the average contribution related information such as birth date, hire date, gender, rate is lower for the post-automatic enrollment cohort and compensation, as well as savings plan information because a sizeable fraction of participants persist at the such as initial plan eligibility and participation dates, (low) default contribution rate of 2% (Madrian and Shea, current participation status, and year-end measures for 2001; Choi, et al., 2002 and 2004; Beshears, et al., total balances, outstanding loans amounts, and asset 2008). Automatic enrollment increases total potential allocation. We also have a monthly contribution rate retirement system balances by 7% of starting pay eight history, as well as annual measures of contributions, years after hire; at the same time, leakage in the form withdrawals and loan payments. The data span calendar of outstanding loans and withdrawals that are not rolled years 2005 through 2013. over into another qualified savings plan also increase by 3% of starting pay, offsetting approximately 40% of the

Potential vs. realized savings under automatic enrollment | July 2018 2 Our analysis focuses on comparing savings outcomes for identify whether withdrawals from the plan come out two cohorts of newly hired employees at the firm studied. of rolled-in balances or from contributions made to The pre-automatic enrollment (pre-AE) cohort consists of plan. The possibility of taking withdrawals from rolled-in employees hired in the year preceding the introduction balances has the potential to generate extreme outliers of automatic enrollment—that is, from July 1, 2004, to in our measures of leakage which, like plan balances, June 30, 2005. The post-automatic enrollment (post-AE) we normalized by starting pay. This further reduces our cohort consists of employees hired in the year following sample by an additional 3% for both cohorts. Our final the introduction of automatic enrollment—that is, from sample includes 14,883 employees, 7,347 in the pre-AE July 1, 2005, to June 30, 2006. Because most of our cohort and 7,536 in the post-AE cohort. data come from cross-sectional year-end snapshots, individuals in the pre-AE cohort have tenures ranging We begin our analysis by documenting the differences from 6-17 months at year-end 2005, as do the post-AE in savings plan outcomes that can be directly observed cohort at year-end 2006. In our analysis, we will, for the in our data for the pre- and post-AE cohorts. These sake of parsimony, label both of these cohorts as having outcomes include: eligibility for, participation in, tenure of one year at these respective points in time, contributions to, balances in, loans from and withdrawals although in fact each cohort will have individuals with from the savings plan. Table 2 reports these outcomes tenures ranging from 6-17 months (one year on average). and various demographic characteristics. We follow a similar convention for subsequent levels of tenure up through eight years. The final observation for III. Key results the pre-AE cohort in our analysis comes from year-end 2012, when this cohort has 90-101 months of tenure, We find that automatic enrollment has two opposing while for the post-AE cohort, that comes from year-end effects on the preservation of retirement assets following 2013, when these employees are at a similar level of separation: conditional on balances at separation, tenure (see Appendix Table 1 for the precise tenure levels leakage rates are higher post-AE, which works to reduce of both cohorts at each year-end spanned by our data). retirement system balances, but automatic enrollment also increases the balances that employees have at Our primary outcome variable of interest is the ratio separation, which tends to reduce leakage. of savings plan balances to starting pay. However, our year-end snapshots only contain salary information for To gauge the total impact of automatic enrollment individuals who are employed on the snapshot date. retirement savings, we need to account for both the Because of this, we exclude from our sample employees higher balances accrued under automatic enrollment and who were hired during the relevant date ranges for our the higher rate of leakage. To do this, we compare four two hire cohorts, but who were not still employed at different measures of imputed balances for the pre- and the firm on the snapshot date corresponding to tenure post-AE cohorts: of one year as defined above and in Appendix Table 1 WW (12/31/2005 for the pre-AE cohort and 12/31/2006 Contribution-inferred potential plan balances: As for the post-AE cohort). Because turnover rates at this described earlier, this is the projected value of plan company are quite high, this selection criterion excludes balances under a common set of assumptions 45% of the pre-AE cohort and 44% of the post-AE cohort. regarding plan eligibility and asset returns over time We additionally exclude any individuals employed at the for both cohorts and assuming there are no loans or tenure year one snapshot date whose salary information withdrawals from the plan. is missing for some reason other than non-employment; WW Contribution-inferred retirement system balances this restriction reduces our sample by an additional 3%. (including loans): This measure adds to contribution- Finally, we also exclude employees who rolled balances inferred plan balances the cumulative projected into the plan because, in our data, we cannot easily value of rollover withdrawals plus the imputed value

Potential vs. realized savings under automatic enrollment | July 2018 3 of outstanding loans (calculated as contribution- experienced by each cohort at any given tenure time. inferred potential plan balances multiplied by (1-L3)).1 Another difference is that the gap in balances as a Outstanding loan balances are treated as if they will be fraction of pay between the pre- and post-AE cohorts is repaid and remain in the retirement savings system. much less variable over time for the contribution-inferred measure of plan balances relative to pay than for actual WW Contribution-inferred retirement system balances plan balances relative to pay. Finally, the differences in (excluding loans): this measure adds to contribution- balances relative to pay between the pre- and post-AE inferred plan balances the cumulative projected value cohorts at higher levels of tenure is much smaller for our of rollover withdrawals (calculated as contribution- contribution-inferred measure of plan balances to pay inferred potential plan balances multiplied by (1-L2)).2 than for actual plan balances to pay. At eight years of Outstanding loan balances are treated in the same tenure, actual plan balances for the post-AE cohort of the way as non-rollover withdrawals. continuously employed are higher by an amount equal to 17% starting pay; in contrast, contribution-inferred WW Contribution-inferred plan balances: Contribution- plan balances for the post-AE, continually employed inferred potential plan balances net of the cumulative cohort are higher by only 4% of starting pay. This smaller projected value of rollover and non-rollover difference reflects two factors. First, the eligibility withdrawals and loans (calculated as contribution- changes that we account for in constructing our measure inferred potential plan balances multiplied by (1-L1)).3 of contribution-inferred balances excludes up to four months of contributions for the post-AE cohorts, reducing Figure 1 shows the evolution of these different balance their accumulation relative to the pre-AE cohort. Second, measures over time for the pre- and post-AE cohorts the post-AE cohort experiences the financial crisis at a of all hires and for the subsample of the continuously lower level of tenure than does the pre-AE cohort. Our employed. Table 3 shows the value of these measures approach to constructing contribution-inferred balances at selected levels of tenure. In both Figure 1 and Table applies the same pre-AE time sequence of asset returns 3, we normalize these measures of contribution-inferred to both cohorts, delaying in tenure time the contributions balances by starting pay. made by the post-AE cohort that experience high market returns as the economy recovers from the stock market One way to assess the impact of our approach to crash of 2008, and compressing the differences in asset calculating contribution-inferred balances is to compare accumulation across cohorts. the contribution-inferred measure of plan balances relative to starting pay in Figure 1d with ratio of actual Comparing our different measures of contribution-inferred plan balances to starting pay in Figure 2. One clear balances in Figure 1, we see that all of the measures of difference is that our measures of contribution-inferred contribution-inferred balances are higher for the post-AE plan balances for the pre- and post-AE cohorts have very cohort than for the pre-AE cohort. As one might expect, similar slopes between any two tenure years, whereas the largest differences are for potential plan balances the measures of actual plan balances relative to pay (which do not account for leakage), which are higher for have slopes between any two tenure years that are the post-AE cohort of all hires by 7.3% of starting pay out of synch because calendar time asset returns are at eight years of tenure. The difference in contribution- experienced by the pre- and post-AE cohorts at different inferred plan balances between the pre- and post-AE points in tenure time. This difference is by design, as cohorts is much smaller, at 3.4% of starting pay for all our contribution-inferred methodology fixes asset returns hires at eight years of tenure, reflecting the

1 L3 = cumulative non-rollover leakage rate 2 L2 = cumulative non-rollover + loan leakage rates 3 L1 = cumulative rollover + cumulative non-rollover + loan leakage rates

Potential vs. realized savings under automatic enrollment | July 2018 4 fact that loans and withdrawals drive a wedge between pre-AE cohort of all hires at eight years of tenure (first potential balances and what actually remains in the plan, row of Table 4), and by a somewhat larger 17.8% for the and that this wedge is larger for the post-AE cohort. But post-AE cohort (second row of Table 4). The higher rate of some plan withdrawals do not reflect leakage from the non-rollover withdrawals for the post-AE cohort reduces retirement system as a whole, just leakage from the the potential savings gains of automatic enrollment at plan. So if our interest is in retirement system balances eight years of tenure by 36.0% (third row of Table 4). If rather than just plan balances, a better metric would be we exclude loans from our measure of retirement system our measures of contribution-inferred retirement system balances, all of these numbers increase: non-rollover balances, which includes the cumulative projected withdrawals decrease potential balances by 16.9% for value of rollover withdrawals. If we include loans in our the pre-AE cohort and by a higher 22.1% for the post-AE measure of retirement system balances, they are higher cohort, reducing the potential savings gains of automatic by 4.6% of starting pay for the post-AE cohort of all hires; enrollment by 41.6%. if we exclude loans from our measure of retirement system balances, they are higher by a slightly smaller If we look at our continuously employed subsample in 4.2% of starting pay. the bottom of panel of Table 4, non-rollover withdrawals offset 9.1% of the potential savings gains of automatic Table 4 shows the proportionate change in contribution- enrollment at eight years of tenure (relative to the 36% inferred retirement system and plan balances relative offset for the all hires sample), while the combination of to our measure of contribution-inferred potential plan non-rollover withdrawals and loans offset 27.4% of the balances. The Pre- and Post-AE rows in Table 4 use the potential savings gains of automatic enrollment. Which of numbers in Table 3 to calculate the fraction of potential these very different offset measures is a more accurate plan balances that are “lost” to either the retirement reflection of the extent to which incremental leakage for system or to the plan due to loans and withdrawals (e.g., the post-AE cohort offsets some of the savings gains of the 5.7% in the first cell in Table 4 is calculated as (5.3- automatic enrollment depends on the extent to which 4.99)/4.99 taken from the pre-AE Potential Plan Balances loans are repaid. As noted earlier, the data from this and Retirement System Balances (incl. loans) rows in Table firm suggests that almost 90% of loan balances are 3). The Difference rows measure the extent to which eventually repaid (Appendix Figure 1), so the smaller loans and withdrawals offset the potential increases number is probably closer to the truth, although with a in savings generated by automatic enrollment. A value downward bias. of 0 indicates that all of the increases in contribution- inferred potential plan balances generated by automatic The analysis thus far shows the impact of automatic enrollment are retained as increased saving (this does enrollment on population average outcomes, either not imply that there is no leakage, just that there is for the population of all hires, or for the continuously no incremental leakage from the increased balances employed subgroup. These means differences mask induced by automatic enrollment), whereas a value considerable heterogeneity in the impact of automatic of 1 indicates that all of increased savings generated enrollment. In Figure 3, we plot contribution-inferred by automatic enrollment are offset by an increase potential plan balances at different points in the in leakage. Numbers between 0 and 1 measure the savings plan distribution for the all hires population. At share of the automatic-enrollment induced increase in eight years of tenure, the average impact of automatic contribution-inferred potential plan balances that are enrollment on contribution-inferred potential plan offset by increased leakage for the post-AE cohort. balances is an increase of 7.2% of starting pay. The impacts at eight years of tenure at the 10th, 25th, 50th, Relative to the level of contribution-inferred potential plan 75th and 90th percentiles of the potential plan balance balances, non-rollover withdrawals decrease contribution- distribution are 1.9%, 4.6%, 8.4%, 10.9% and 6.4% inferred retirement system balances by 13.0% for the respectively. As we have documented, much of this

Potential vs. realized savings under automatic enrollment | July 2018 5 increase in potential plan balances is never realized, so so for individuals who separate with longer tenures. The we look in Figure 4 at the distribution of contribution- effects at the higher percentiles of the distribution are inferred plan balances (as opposed to potential plan much smaller and not statistically different from 0 for balances). The average impact of automatic enrollment employees who separate with higher levels of tenure (the on contribution-inferred plan balances at eight years sample sizes of employees separating with higher levels of tenure for all hires is an increase of 3.4% of starting of tenure are relatively small). pay; the impacts at the 10th, 25th, 50th, 75th and 90th percentiles are 0%, 0%, 0%, 27.5% and 6.7%. Because In Figure 8, we measure how much of the increase in turnover at this firm is high, the lower end of the savings potential plan balances shown in Figure 7 is realized by distribution is primarily composed of employees who plotting the difference in contribution-inferred retirement have separated, while the upper end of the savings system balances (excluding loans) for the pre- and distribution is primarily composed of those who have post-AE cohorts (the y-axes plot the post minus the been continuously employed. The complete lack of an pre-AE outcomes). There is no difference at the 10th impact at the 10th, 25th, and 50th percentiles results percentile of the distribution; all of the balances of from the level of contribution-inferred plan balances both cohorts are withdrawn and leave the retirement to starting pay being 0 for both the pre- and post-AE system after separation. There is also no difference cohorts. Essentially, the individuals at these points in the at the 25th percentile except for a small effect among distribution have separated from the firm and taken all of employees who separate with seven years of tenure. their balances out of the plan. We don’t see a sustained The place in the distribution where we see an effect positive impact of automatic enrollment on plan balances of automatic enrollment is at the median. The effect until the 75th percentile of the distribution. Although the on retirement system balances increases with tenure measure of contribution-inferred balances in Figure 4 at time of separation through five years of tenure excludes rollovers, it illustrates the significant impact that and then declines. This pattern is consistent with the withdrawals have on both the pre- and post-AE cohorts. data presented in Figure 9 on asset preservation. The separated participant at the median of the savings Figures 5 and 6 show the same outcomes as Figures 20 distribution has accumulated some balances in the plan, and 21 for the subsample of the continuously employed. and automatic enrollment has the effect of increasing The levels of contribution-inferred potential plan balances balances at separation enough to move some of these are much higher for both the pre- and post-AE cohorts participants across the balances-at-separation categories compared to the levels for the sample of all hires. The in Figure 9 in a way that preserves assets, reducing the distributional effects of automatic enrollment are largest fraction of participants who are subject to a compelled at the 10th and 25th percentiles, and there is very little cash distribution (balances at separation <$1,000) and effect at higher percentiles in the savings distribution. increasing the fraction for whom the default is an IRA rollover or the fraction who can keep their balances in In Figures 7 and 8, we examine the distributional the plan. The effect at the median likely decreases with outcomes for the subsample of employees who have tenure at separation at higher levels of tenure because separated from the firm. Instead of showing how savings as tenure increases, balances in the plan also increase evolves with tenure, the outcomes in these figures and are more likely to stay in the retirement system are all measured eight years after hire, but employees after separation for both the pre- and post-AE cohorts. are stratified by their tenure at the time of separation Consistent with this, at the 75th and 90th percentile, (the x-axes). In Figure 7, we measure the difference in the differences in retirement system balances across contribution-inferred potential plan balances for the and pre- and post-AE cohorts are very small and/ pre- and post-AE cohorts (the y-axes plot the post minus or not statistically different from 0. These individuals the pre-AE outcomes). Automatic enrollment increases are the motivated savers and likely to preserve assets contribution-inferred potential plan balances at the 10th regardless of balances at the time of separation. and 25th percentiles of the savings distribution, more

Potential vs. realized savings under automatic enrollment | July 2018 6 IV. Conclusion those who separate. Among those who remain employed, leakage offsets relatively little of the incremental savings Our analysis highlights the potential magnitude that generated by automatic enrollment at low levels of pre-retirement withdrawals and loans have on retirement tenure. As tenure increases, so does the extent to which system balances in general, and in attenuating the leakage offsets the savings increases from automatic potential impact of automatic enrollment on asset enrollment, and eight years after hire, leakage, primarily accumulation in particular. We find that automatic in the form of plan loans, offsets 9-27% of the potential enrollment increases total potential retirement system increased savings. In contrast, for employees who balances by 7% of starting pay eight years after hire; at separate, leakage, primarily in the form of non-rollover the same time, leakage in the form of outstanding loans withdrawals, offsets more than half of the potential and withdrawals that are not rolled over into another incremental savings from automatic enrollment at low qualified savings plan also increase by 3% of starting levels of tenure. Although this rate of offset declines with pay, offsetting approximately 40% of the potential time since hire for separated employees, at eight years increase in savings from automatic enrollment. The net it still exceeds 40%. Overall, while automatic enrollment effect is that automatic enrollment increases retirement results in a net increase in retirement system balances, system balances by 4-5% of first year pay eight years pre-retirement leakage significantly limits its potential after hire. These results mask substantial differences impact. across those who remain employed at the firm versus

Potential vs. realized savings under automatic enrollment | July 2018 7 References

Argento, Robert, Victoria L. Bryant, and John Sabelhaus, 2015. “Early Withdrawals from Retirement Accounts during the Great Recession.” Contemporary Economic Policy 33(1), 1-16. Beshears, John, James J. Choi, David Laibson, and Brigitte C. Madrian. 2008. “The Importance of Default Options for Retirement Saving Outcomes: Evidence from the United States.” In Stephen J. Kay and Tapen Sinha, eds., Lessons from Pensions Reform in the Americas, Oxford: Oxford University Press, 59-87. Choi, James J., David Laibson, Brigitte C. Madrian, and Andrew Metrick, 2002. “Defined Contribution Pensions: Plan Rules, Participant Decisions, and the Path of Least Resistance.” In James Poterba, ed., Tax Policy and the Economy, Vol. 16, 67-114. Choi, James J., David Laibson, Brigitte C. Madrian, and Andrew Metrick, 2004. “For Better or For Worse: Default Effects and 401(k) Savings Behavior.” In David A. Wise, ed., Perspectives on the Economics of Aging, Chicago: Chicago University Press, 81-126. Madrian, Brigitte C. and Dennis F. Shea, 2001. “The Power of Suggestion: Inertia in 401(k) Participation and Savings Behavior.” Quarterly Journal of Economics 116(4), 1149-1187.

Potential vs. realized savings under automatic enrollment | July 2018 8 About the authors

John Beshears is the Terrie F. and Bradley M. Bloom Associate Professor of Business Administration in the Negotiation, Organizations & Markets Unit of Harvard Business School. He also serves as a faculty research fellow at the National Bureau of Economic Research. Before joining HBS, he was an assistant professor of finance at the Stanford Graduate School of Business.

Professor Beshears’s primary area of research is behavioral economics, which combines insights from psychology and economics to explore individual decision making and market outcomes.

His work has been published in the Journal of Finance, the Journal of Financial Economics, Review of Financial Studies, the Journal of Public Economics, the Journal of Economic Behavior & Organization and Proceedings of the National Academy of Sciences of the United States of America. Additional work has been featured in The Economist, The Wall Street Journal, The New York Times, BusinessWeek and Time.

After earning his Ph.D. in business economics at HBS, Professor Beshears was a postdoctoral fellow at the National

James J. Choi is Professor of Finance at the Yale School of Management. His research addresses issues spanning behavioral finance, behavioral economics, household finance, capital markets, health economics, and sociology. His work on default options has led to changes in 401(k) plan design at many U.S. corporations and has influenced pension legislation in the U.S. and abroad.

Choi’s research also has examined the influence of racial, gender, and religious identity on economic preferences, investor ignorance of mutual fund fees, the effect of deadlines and peer information on savings choices, and how retail investor sentiment in China affects stock returns. He earned his A.B. in applied mathematics and Ph.D. in economics from Harvard University.

David Laibson is the Robert I. Goldman Professor of Economics. He leads Harvard University’s Foundations of Human Behavior Initiative. Laibson’s research focuses on the topic of behavioral economics, with emphasis on household finance, macroeconomics, aging, and intertemporal choice. Laibson is also a member of the National Bureau of Economic Research, where he directs the National Institute of Aging Roybal Center for Behavior Change in Health and Savings, and is a Research Associate in the Aging, Asset Pricing, and Economic Fluctuations Working Groups. Laibson serves on the Board of the Russell Sage Foundation and on Harvard’s Pension Investment Committee. Laibson serves on the advisory board of the Social Science Genetics Association Consortium. Laibson has served as the Chair of the Department of Economics at Harvard University and as member of the Academic Research Council of the Consumer Financial Protection Bureau. Laibson is a recipient of a Marshall Scholarship. He is a Fellow of the Econometric Society and the American Academy of Arts and Sciences. He is a two-time recipient of the TIAA-CREF Paul A. Samuelson Award for Outstanding Scholarly Writing on Lifelong Financial Security. Laibson holds degrees from Harvard University (AB in Economics, Summa), the London School of Economic (MSc in Econometrics and Mathematical Economics), and the Massachusetts Institute of Technology (Ph.D. in Economics). He received his PhD in 1994 and has taught at Harvard since then. In recognition of his teaching, he has been awarded Harvard’s Prize and a Harvard College Professorship. ΦΒΚ

Potential vs. realized savings under automatic enrollment | July 2018 9 Brigitte Madrian is the Aetna Professor of Public Policy and Corporate Management at the Harvard Kennedy School. Before coming to Harvard in 2006, she was on the faculty at the University of Pennsylvania Wharton School (2003- 2006), the University of Chicago Graduate School of Business (1995-2003) and the Harvard University Economics Department (1993-1995). She also is a research associate and co-director of the Household Finance working group at the National Bureau of Economic Research.

Madrian’s current research focuses on behavioral economics and household finance, with a particular focus on household saving and investment behavior. Her work in this area has impacted the design of employer-sponsored savings plans in the U.S. and has influenced pension-reform legislation both in the U.S. and abroad. She also is engaged in research on health, using the lens of behavioral economics to understand health behaviors and improve health outcomes.

Madrian received her Ph.D. in economics from the Massachusetts Institute of Technology and studied economics as an undergraduate at Brigham Young University. She is the recipient of the National Academy of Social Insurance Dissertation Prize (first place, 1994) and a two-time recipient of the TIAA Paul A. Samuelson Award for Scholarly Research on Lifelong Financial Security (2002 and 2011).

Potential vs. realized savings under automatic enrollment | July 2018 10 Table 1. Retirement savings plan features Eligibility Employee contributions Before 7/1/2005: First day of the month following three full months of continuous service for employees scheduled to work 20+ hours per week On or after 7/1/2005: Immediately upon hire for all employees

Employer contributions First day of the month following one year of service in which the employee worked 1,000+ hours and if employed at the end of the year Automatic Enrollment Employees hired on or after July 1, 2005, are automatically enrolled in the plan at a 2% contribution rate invested in a balanced mutual fund unless they opt-out within five business days Automatic Escalation Available as an opt-in feature starting August 1, 2006, with contribution escalation occurring on January 1 of each subsequent calendar year Contributions

Employee Before 1/1/2006: up to 50% of pay On or after 1/1/2006: up to 75% of pay Employer 100% match on employee contributions up to 4% of pay, allocated to employer stock Vesting Immediate Loans Total loan limit At most two loans outstanding at a time Loan minimum $1,000 Loan maximum The lesser of 50% of the participant’s account balance or $50,000 minus the participant’s highest outstanding loan balance during the past 12 months Distributions Balances <$1,000 are subject to an automatic cash distribution if not rolled into another qualified plan within 60 days. following separation Balances of $1,000 to <$5,000 are automatically rolled into an IRA if not rolled into another qualified plan or taken as a cash distribution. Balances >$5,000 can be retained in the plan after separation, rolled into another qualified plan, or taken as a cash distribution. Distributions taken before age 55 and not rolled into another qualified account are subject to 10% tax penalty.

In-service withdrawals Non-Hardship Permitted from all accounts after age 59½ without penalty and from after-tax and certain rollover accounts before age 59½ with a 10% penalty Hardship Permitted from all accounts for college, funeral, outstanding medical, and some primary residence expenses without penalty; $500 minimum Source: Plan documents.

Potential vs. realized savings under automatic enrollment | July 2018 11 Table 2. Demographic characteristics and savings plan outcomes Post-AE p-value of Pre-AE cohort cohort difference Demographic characteristics Fraction female 64.5% 65.7% 0.119 Age at hire 31.1 31.1 0.938 Avg. starting salary ($2004)a $28,551 $28,285 0.450 Months to eligibility from hire 3.5 0.1 0.000 Ever contributed to savings plan 62.2% 98.3% 0.000 Median months to participation from eligibility | ever contributed 5 0 0.000 Participation rate in first month of eligibility 20.4% 77.1% 0.000 Continuously employed as of eight years after hire 14.8% 14.8% 0.895 Savings plan outcomes (at one year after hire) Participation rate | still employed 36.2% 96.0% 0.000 Avg. contribution rate | still employed 1.7% 3.0% 0.000 Avg. contribution rate | contributing and still employed 5.8% 3.3% 0.000 Balance/starting salary | still employed 1.4% 3.2% 0.000 Savings plan outcomes (at eight years after hire) Participation rate | still employed 86.6% 96.0% 0.000 Avg. contribution rate | still employed 5.8% 6.0% 0.581 Avg. contribution rate | contributing and still employed 7.1% 6.4% 0.046 Balance/starting salary | still employed 18.7% 25.4% 0.000 Fraction with outstanding loan | still employed and participating 26.3% 31.5% 0.008 Number of loans | loans and still employed 1.63 1.61 0.673 Loan balance/starting salary | loan and still employed 19.1% 19.1% 0.999 Ever taken rollover withdrawal | ever contributed 28.8% 25.3% 0.000 Ever taken non-rollover withdrawal | ever contributed 43.2% 58.6% 0.000 Sample size N=7,347 N=7,536

Source: Authors’ calculations. The sample is all hires continuously employed through eligible for the plan at tenure year one as defined in Appendix Table 1. a Growth in seasonally adjusted average weekly earnings for private sector workers from the Current Employment Statistics survey is used to deflate employee salaries to 2004 dollars.

Potential vs. realized savings under automatic enrollment | July 2018 12 Table 3. Contribution-inferred balances relative to starting pay A. All Hires Tenure (years) 2 4 6 8 Potential Plan Balances Pre-AE 5.30% 9.02% 18.55% 27.39% Post-AE 8.65% 12.51% 24.02% 34.63% Difference (Post-Pre) 3.36% 3.49% 5.47% 7.24% Retirement System Balances (incl. loans) Pre-AE 4.99% 8.15% 16.45% 23.83% Post-AE 7.78% 10.65% 20.06% 28.46% Difference (Post-Pre) 2.78% 2.50% 3.61% 4.64% Retirement System Balances (excl. loans) Pre-AE 4.95% 7.90% 15.78% 22.75% Post-AE 7.70% 10.29% 19.11% 26.98% Difference (Post-Pre) 2.75% 2.39% 3.33% 4.23% Plan Balances Pre-AE 4.60% 6.75% 13.46% 18.48% Post-AE 7.17% 8.87% 16.10% 21.88% Difference (Post-Pre) 2.57% 2.11% 2.64% 3.39% B. Continuously Employed Tenure (years) 2 4 6 8 Potential Plan Balances Pre-AE 7.21% 18.93% 48.19% 82.45% Post-AE 11.05% 22.34% 52.85% 88.16% Difference (Post-Pre) 3.84% 3.41% 4.66% 5.71% Retirement System Balances (incl. loans) Pre-AE 7.19% 18.77% 47.66% 81.44% Post-AE 10.97% 22.08% 51.85% 86.63% Difference (Post-Pre) 3.78% 3.31% 4.19% 5.19% Retirement System Balances (excl. loans) Pre-AE 7.11% 18.10% 45.49% 77.32% Post-AE 10.84% 21.22% 48.83% 81.46% Difference (Post-Pre) 3.73% 3.13% 3.34% 4.14% Plan Balances Pre-AE 7.11% 18.10% 45.49% 77.03% Post-AE 10.84% 21.22% 48.71% 81.36% Difference (Post-Pre) 3.74% 3.12% 3.22% 4.33%

Source: Authors’ calculations. The sample is all hires continuously employed through and eligible for the plan at tenure year one as defined in Appendix Table 1 (panel A). The continuously employed subsample is restricted to those continuously employed through and eligible for the plan at the indicated level of tenure (panel B). The different measures of contribution-inferred balances are calculated as described in Section III of this report.

Potential vs. realized savings under automatic enrollment | July 2018 13 Table 4. Reduction in balances relative to contribution-inferred potential plan balances A. All Hires Tenure (years) 2 4 6 8 Contribution-Inferred Retirement System Balances (incl. loans) Pre-AE -5.7% -9.6% -11.3% -13.0% Post-AE -10.1% -14.8% -16.5% -17.8% Differential Impact (Post-Pre) 17.1% 28.3% 34.0% 36.0% Contribution-Inferred Retirement System Balances (excl. loans) Pre-AE -6.6% -12.4% -14.9% -16.9% Post-AE -11.0% -17.8% -20.4% -22.1% Differential Impact (Post-Pre) 18.0% 31.6% 39.1% 41.6% Contribution-Inferred Plan Balances Pre-AE -13.2% -25.1% -27.4% -32.5% Post-AE -17.2% -29.1% -33.0% -36.8% Differential Impact (Post-Pre) 23.4% 39.4% 51.7% 53.2% B. Continuously Employed Tenure (years) 2 4 6 8 Contribution-Inferred Retirement System Balances (incl. loans) Pre-AE -0.3% -0.9% -1.1% -1.2% Post-AE -0.7% -1.2% -1.9% -1.7% Difference (Post-Pre) 1.6% 2.9% 10.1% 9.1% Contribution-Inferred Retirement System Balances (excl. loans) Pre-AE -1.4% -4.4% -5.6% -6.2% Post-AE -1.9% -5.0% -7.6% -7.6% Difference (Post-Pre) 2.9% 8.3% 28.3% 27.4% Contribution-Inferred Plan Balances Pre-AE -1.4% -4.4% -5.6% -6.6% Post-AE -1.9% -5.0% -7.8% -7.7% Difference (Post-Pre) 2.8% 8.4% 30.8% 24.2%

Potential vs. realized savings under automatic enrollment | July 2018 14 Figure 1. Contribution-inferred balances relative to starting pay (person-weighted)

Source: Authors’ calculations. The sample is all hires continuously employed through and eligible for the plan at tenure year one as defined in Appendix Table 1 (panel A). The “employed” subsample is restricted to those continuously employed through and eligible for the plan at the indicated level of tenure (panel B). The different measures of contribution-inferred balances are calculated as described in Section III of this report. Starting pay is the annualized salary during the calendar year corresponding to tenure year.

Potential vs. realized savings under automatic enrollment | July 2018 15 Figure 2. Savings plan balances relative to starting pay

Source: Authors’ calculations. The sample is all hires continuously employed through tenure year one as defined in Appendix Table 1. The “employed” subsample is restricted to those continuously employed through and eligible for the plan at the indicated level of tenure. Plan balances include before-tax, after-tax, Roth and employer match balances. Outstanding 401(k) loan amounts are excluded. Starting pay is the annualized salary during the calendar year corresponding to tenure year one. Person-weighted ratio.

Potential vs. realized savings under automatic enrollment | July 2018 16 Figure 3. Contribution-inferred potential plan balances relative to starting pay (all hires)

Source: Authors’ calculations. The sample is all hires continuously employed through and eligible for the plan at tenure year one as defined in Appendix Table 1. Contribution-inferred potential plan balances are calculated as described in Section III of this report. Starting pay is the annualized salary during the calendar year corresponding to tenure year.

Potential vs. realized savings under automatic enrollment | July 2018 17 Figure 4. Contribution-inferred plan balances relative to starting pay (all hires)

Source: Authors’ calculations. The sample is all hires continuously employed through and eligible for the plan at tenure year one as defined in Appendix Table 1. Contribution-inferred plan balances are calculated as described in Section III of this report. Starting pay is the annualized salary during the calendar year corresponding to tenure year.

Potential vs. realized savings under automatic enrollment | July 2018 18 Figure 5. Contribution-inferred potential plan balances to starting pay (continuously employed)

Source: Authors’ calculations. The sample is restricted to those continuously employed through and eligible for the plan at the indicated level of tenure as defined in Appendix Table 1. Contribution-inferred potential plan balances are calculated as described in Section III of this report. Starting pay is the annualized salary during the calendar year corresponding to tenure year.

Potential vs. realized savings under automatic enrollment | July 2018 19 Figure 6. Contribution-inferred plan balances to starting pay (continuously employed)

Source: Authors’ calculations. The sample is restricted to those continuously employed through and eligible for the plan at the indicated level of tenure as defined in Appendix Table 1. Contribution-inferred plan balances are calculated as described in Section III of this report. Starting pay is the annualized salary during the calendar year corresponding to tenure year.

Potential vs. realized savings under automatic enrollment | July 2018 20 Figure 7. Post – pre-AE contribution-inferred potential plan balances relative to starting pay eight years after hire by year of separation (separated employees)

Source: Authors’ calculations. The sample is all hires continuously employed through tenure year one (as defined in Appendix Table 1) who subsequently separated from the firm prior to tenure year eight. Contribution-inferred potential plan balances are calculated as described in Section III of this report and measured at tenure year eight. Starting pay is the annualized salary during the calendar year corresponding to tenure year. 95% confidence interval is included.

Potential vs. realized savings under automatic enrollment | July 2018 21 Figure 8. Post – pre-AE contribution-inferred retirement system balances (excluding loans) relative to starting pay eight years after hire by year of separation (separated employees)

Source: Authors’ calculations. The sample is all hires continuously employed through tenure year one (as defined in Appendix Table 1) who subsequently separated from the firm prior to tenure year eight. Contribution-inferred retirement system balances are calculated as described in Section III of this report and measured at tenure year eight. Starting pay is the annualized salary during the calendar year corresponding to tenure year. 95% confidence interval is included.

Potential vs. realized savings under automatic enrollment | July 2018 22 Figure 9. Asset preservation following separation by imputed balances at separation

Source: Authors’ calculations. The sample is all hires continuously employed through tenure year one (as defined in Appendix Table 1) who subsequently separated from the firm prior to tenure year 8. Imputed balances at separation are calculated as described in Section III of this report. Preserving assets following separation is defined as taking a rollover withdrawal between separation and the year-end following the year of separation or having positive plan balances at the year-end following the year of separation.

Potential vs. realized savings under automatic enrollment | July 2018 23 Appendix Table 1. Tenure Levels of the pre- and post-AE cohorts at different points in calendar time Pre-AE Cohort Post-AE Cohort Hired 7/1/2004 to 6/30/2005 Hired 7/1/2005 to 6/30/2006 Date of year-end Tenure Tenure Tenure Tenure data observation label range label range 12/31/2005 Year 1 6-17 months N/A N/A 12/31/2006 Year 2 18-29 months Year 1 6-17 months 12/31/2007 Year 3 30-41 months Year 2 18-29 months 12/31/2008 Year 4 42-53 months Year 3 30-41 months 12/31/2009 Year 5 54-65 months Year 4 42-53 months 12/31/2010 Year 6 66-77 months Year 5 54-65 months 12/31/2011 Year 7 78-89 months Year 6 66-77 months 12/31/2012 Year 8 90-101 months Year 7 78-89 months 12/31/2013 N/A N/A Year 8 90-101 months

Potential vs. realized savings under automatic enrollment | July 2018 24 Appendix Figure 1. Loan repayment and default as a fraction of cumulative loan amounts

Source: Authors’ calculations. The sample is restricted to those continuously employed through and eligible for the plans at tenure year one (as defined in Appendix Table 1) who have ever participated in the savings plan and taken out a plan loan. The figure shows the fraction of cumulative loan amounts ever borrowed that are active (still outstanding), have been repaid, or have been closed without being repaid (default) at the indicated level of tenure.

Potential vs. realized savings under automatic enrollment | July 2018 25